Portals that work, not just go live
An Experience Cloud portal that deflects cost and converts — not just one that launches
Most Experience Cloud projects ship a good-looking portal and stop there. We treat the customer, partner, or member site as your highest-traffic relationship surface — the place an AI agent can answer, resolve, and convert in front of your customer — and we stay accountable for whether it moves a number. The data and sharing model under the login decide whether any of that is trustworthy, so that's where we start.
A portal is your highest-leverage surface, not a website project
When customers, partners, or members log in, that's where most of your relationship traffic actually lands: cases, orders, renewals, claims, applications, channel deals. So it's the single best place to either take cost out or pull revenue forward. The expensive mistake is scoping it as a branding-and-layout exercise. The question a buyer should ask isn't 'does it look on-brand' — it's 'what does every authenticated session do for the business, and can I see that in a number.'
The part the marketing skips: a portal is only as smart as the data behind the login
Experience Cloud doesn't get its own data model. It inherits your external org-wide defaults, your account-contact relationships, your record ownership, and your data quality — and external users see records through that, not through a UI you design later. A self-service site that surfaces a stale entitlement, the wrong case, or a partner's view of accounts they shouldn't touch isn't a design bug; it's a data-architecture problem wearing a UI. The moment you put an AI agent on that portal, the agent inherits the same constraint: it can only answer from the data it can reach and the permissions it must respect. Get that layer right first, because it decides whether anything built on top is trustworthy.
- External OWDs set to Private, then opened deliberately with sharing sets and sharing rules — so external users see exactly what they should and nothing more
- Clean account-contact and partner-role relationships, so the right person sees the right records and roll-ups don't leak across accounts
- Entitlements, knowledge, and case data structured so deflection and agent answers are correct, not just plausible
- Guest-user and unauthenticated access locked down separately — the most common source of accidental exposure on a public-facing site
Where Agentforce changes the math for a portal buyer
The reason to do Experience Cloud well right now is that the portal is where an Agentforce agent meets your customer directly, inside the same session and the same permission context as the logged-in user. A grounded agent can deflect a tier-1 case, walk a partner through registering a deal, or guide a member through an application — at 2 a.m., in any language, without a queue. That's real cost removed and revenue pulled forward, and it's measurable. But an agent that invents an entitlement or quotes a wrong policy on a customer-facing portal is a liability, not a feature. The discipline is the same as the data work: ground it on your records and knowledge, constrain what it's allowed to say, test it against real cases, and watch it in production.
- Case deflection and self-service resolution on a support portal
- Partner deal registration and onboarding guidance on a channel portal
- Guided applications, renewals, and account servicing on member and customer portals
We build it, run it, and tie our fee to the result
Most Experience Cloud consultants hand you a portal at go-live and are gone before anyone knows whether it actually deflects cases or lifts conversion. We do the opposite. We advise, build, then run the portal and the agents on it — tuning grounding, fixing the answers that go wrong, and reporting against the metric you named when you signed. A portion of our fee is tied to that outcome. That alignment is the whole reason to hire a firm instead of a body shop: the people who built it are still on the hook when it has to perform.
How an Experience Cloud engagement runs with us
We sequence the work with our Data-to-Agent method so the data isn't a scramble at launch. Each phase has a clear exit before the next begins.
- Agent Ready: data quality, sharing model, knowledge, and entitlements made fit for an external audience and an agent
- Agent Launch: portal live with one grounded, tightly scoped agent use case — proven before it's widened
- Agent Scale: more personas, journeys, and channels (customer, partner, member)
- Agent Care: monitoring, answer tuning, and accountability for the metric — because an unwatched portal quietly rots
How to think about the number before you build
Pick one metric first. Say a support portal handles a known monthly volume of tier-1 cases; decide up front what share a grounded self-service agent should resolve. That share is deflected cost and agent-hours freed every month, compounding — and it's also the line you'll hold us to. We'd model it against your real volumes and case mix, not borrow someone else's percentage, because a deflection rate that's true for one knowledge base is fiction for another. The point of the exercise is to define what 'good' means before a single component is built, so the portal is designed backward from the outcome instead of forward from a template. The /roi calculator is a fast way to put a first number on it.
Frequently asked
What does a Salesforce Experience Cloud consultant actually do beyond building the portal?
A good one designs the portal backward from a business outcome — deflected cases, faster partner onboarding, higher self-service conversion — and gets the data and sharing model right so the site is trustworthy. Building the UI is the easy part; the external sharing architecture and what happens after go-live is where the value and the risk live. We go further and run the portal after launch, including any AI agents on it, with part of our fee tied to the result.
Should I put an Agentforce agent on my Experience Cloud portal, or is it too risky?
It's worth it when the agent is grounded in correct data and constrained to what it should say, and genuinely risky when it isn't — a customer-facing portal is no place for an agent that guesses at entitlements or policy. We scope a narrow, high-value first use case, ground it on your real knowledge and records, test it against actual cases, and monitor it in production before widening. Done that way it deflects cost and improves the experience; done carelessly it creates liability you'll see in writing.
How long does an Experience Cloud implementation take?
It depends far more on the state of your data and sharing model than on the portal build. If account-contact relationships, entitlements, and knowledge are clean, a focused first portal and use case can go live quickly. If they aren't, that cleanup is the real project — and skipping it is exactly how portals end up surfacing the wrong records to the wrong users. We scope it honestly after looking at your org, not before.
What's the difference between a customer, partner, and member portal in Experience Cloud?
They differ mostly in who logs in and what they're allowed to see and do, which is governed by license type and your sharing model. Customer sites center on self-service support and account management; partner sites add deal registration, lead sharing, and channel data with tighter access control and account roll-ups; member and community sites serve applications, renewals, and engagement. The data-access rules differ for each, which is precisely why the sharing architecture has to be deliberate rather than copied across personas.
How do you make sure external users can't see data they shouldn't?
Through the sharing model: external org-wide defaults set to Private, then opened with sharing sets, sharing rules, and account-based access designed and tested for your specific personas — plus locking down guest and unauthenticated access separately. This is the single most error-prone part of any external-facing Salesforce site, and it gets more critical the moment an AI agent is reaching into records on a user's behalf. We treat it as a core deliverable with explicit test cases, not an afterthought.
Ready when you are
Worth a
conversation?
Tell us one number you'd like AI to move. We'll show you how we'd do it, what it's worth, and how we'd tie our fee to getting you there.